Odyssey is an expanded re-work of Ivory. It is a very versatile and competent model.
Semi-realistic anime style with good contrast and bright colors.
Some assets attained from MidJourney enable this model to produce fantastic landscapes/objects as well.
Works well with LORA/textual inversions.
For best results, generate at (512,768) or (768,512) resolutions.
NSFW capable although far from a strictly hentai model.
All sample images generated using Euler A sampling method at 20 steps.
Upscaled using Foolhardy Remacri at 0.3 denoising strength.
Images are not cherry picked to show true performance.
Description
FP32 Pruned. No baked VAE.
FAQ
Comments (6)
What's the difference between fp16 & fp32 models? I'm seeing this more & more often, so I'm a little confused.
@EvilFear Yeah, I have Googled it. And the only thing that comes up are long complicated explanations about the technical side of AI training that fp16 is a more lightweight training method than fp32. But that doesn't explain in laymen's terms what the difference is in using either model for Image Generation does. Like does it affect quality of output? Is it quicker and lighter on hardware that isn't above a certain range? Nothing I've found has said what the pros and cons are in regard to Stable Diffusion. SO, I asked a creator who obviously might know. You obtuse troll.
@EvilFear wow you must be a fun dood
The simple explanation is that FP16 is less precise than FP32. FP16 uses half the amount of bits compared to FP32. For generating images with stable diffusion this loss of precision is negligible. So using the FP16 version will save you disk space and VRAM for basically no loss of quality.
@laucx Thank you! <3 <3